-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathtrainer.py
182 lines (125 loc) · 5.35 KB
/
trainer.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
import os
from tqdm import tqdm
from utils.data import get_dataloader
import config.const as const_util
from utils import Context as ctxt
from tester import Tester
# from tester_ovo import Tester
import torch
import torch.optim as optim
import logging
class Trainer(object):
def __init__(self, flags_obj, cm, dm, new_config=None):
"""
Args:
flags_obj: arguments in main.py
cm : context manager
dm : dataset manager
new config : update default model config(`./config/model_kuaishou.yaml`) to tune hyper-parameters
"""
self.name = flags_obj.name + '_trainer'
self.cm = cm #context manager
self.dm = dm #dataset manager
self.flags_obj = flags_obj
self.set_recommender(flags_obj, cm.workspace, dm, new_config)
self.recommender.transfer_model()
self.lr = self.recommender.model_config['lr']
self.tester = Tester(flags_obj, self.recommender)
def set_recommender(self, flags_obj, workspace, dm, new_config):
self.recommender = ctxt.ContextManager.set_recommender(flags_obj, workspace, dm, new_config)
def train(self):
self.set_dataloader()
self.set_optimizer()
self.set_scheduler()
self.set_esm() #early stop manager
best_metric = 0
train_loss = [0.0, 0.0, 0.0, 0.0] #store every training loss
val_loss = [0.0]
for epoch in range(self.flags_obj.epochs):
print('epoch:{}'.format(epoch))
self.train_one_epoch(epoch, train_loss)
watch_metric_value = self.validate(epoch, val_loss)
if watch_metric_value > best_metric:
self.recommender.save_ckpt()
logging.info('save ckpt at epoch {}'.format(epoch))
best_metric = watch_metric_value
self.scheduler.step(watch_metric_value)
stop = self.esm.step(self.lr, watch_metric_value)
if stop:
break
def set_test_dataloader(self):
raise NotImplementedError
def test(self, assigned_model_path = None, load_config=True):
'''
test model on test dataset
Args:
tune_para
'''
self.set_test_dataloader()
if load_config:
self.recommender.load_ckpt(assigned_path = assigned_model_path)
results = self.tester.test()
logging.info('TEST results :')
self.record_metrics('test', results)
print('test: ', results)
def set_dataloader(self):
raise NotImplementedError
def set_optimizer(self):
self.optimizer = self.recommender.get_optimizer()
def set_scheduler(self):
self.scheduler = optim.lr_scheduler.ReduceLROnPlateau(self.optimizer,
mode='max', patience=self.recommender.model_config['patience'],
min_lr=self.recommender.model_config['min_lr'])
def set_esm(self):
self.esm = ctxt.EarlyStopManager(self.recommender.model_config)
def record_metrics(self, epoch, metric):
"""
record metrics after each epoch
"""
logging.info('VALIDATION epoch: {}, results: {}'.format(epoch, metric))
def train_one_epoch(self, epoch, train_loss):
self.lr = self.train_one_epoch_core(epoch, self.dataloader, self.optimizer, self.lr, train_loss)
def train_one_epoch_core(self, epoch, dataloader, optimizer, lr, train_loss):
epoch_loss = train_loss[0]
self.recommender.model.train()
current_lr = optimizer.param_groups[0]['lr']
if current_lr < lr:
lr = current_lr
logging.info('reducing learning rate!')
logging.info('learning rate : {}'.format(lr))
tqdm_ = tqdm(iterable=dataloader, mininterval=1, ncols=100)
for step, sample in enumerate(tqdm_):
optimizer.zero_grad()
loss = self.recommender.get_loss(sample,epoch)
loss.backward()
optimizer.step()
epoch_loss += loss.item()
print("epoch {:d} , step {:d} , loss: {:.4f}".format(epoch+1, step+1, epoch_loss / (step+1+epoch*dataloader.__len__())))
logging.info('epoch {}: loss = {}'.format(epoch, epoch_loss/(step+1+epoch*dataloader.__len__())))
train_loss[0] = epoch_loss
return lr
def validate(self, epoch, total_loss):
results = self.tester.test()
self.record_metrics(epoch, results)
print(results)
return results['mrr']
class SAQRec_Trainer(Trainer):
def __init__(self, flags_obj, cm, dm, nc):
super().__init__(flags_obj, cm, dm, nc)
def set_dataloader(self):
# training dataloader
dst = self.recommender.get_dataset(const_util.train_file, self.dm, True)
self.dataloader = get_dataloader(
data_set = dst,
bs = self.dm.batch_size,
prefetch_factor = self.dm.batch_size // 32 + 1, num_workers = 32,
)
# validation dataloader
self.tester.set_dataloader(
dst = self.recommender.get_dataset(const_util.valid_file, self.dm, False), #这里和train的区别是第三项是false,is_train区别
)
def set_test_dataloader(self):
dst = self.recommender.get_dataset(const_util.test_file, self.dm, False)
self.tester.set_dataloader(
dst = dst,
)